Pharmacy benefit management machine learning systems and methods

Patent metadata

Field	Value
Publication number	US-11514528-B1
Application number	US-201916446825-A
Country	US
Kind code	B1
Filing date	Jun 20, 2019
Priority date	Jun 20, 2019
Publication date	Nov 29, 2022
Grant date	Nov 29, 2022

Abstract

A machine learning process for use with a pharmacy benefits management system. The machine learning process identifies a first predicted set of drug benefit claims impacted by a pricing error, reprices a sample of the first predicted set of drug benefit claims to adjust for the error, and trains a predictive model as a function of the repriced sample. Based on the trained model, the machine learning process predicts a second predicted set of drug benefit claims impacted by the error and initiates automatic repricing.

First claim

What is claimed is:

1. A pharmacy benefits management system comprising:
a data store storing pricing data for a plurality of drug benefit claims;
a front end executing on a processor of a benefit manager device, the front end receiving and responsive to user input for generating an adjustment request associated with at least one of the plurality of drug benefit claims;
a database service executing on a processor of the benefit manager device and coupled to the front end, the database service comprising a model repository;
a modeling processor coupled to the data store and to the database service; and
a memory storing computer-executable instructions that, when executed by the modeling processor, configure the modeling processor for:
retrieving, in response to the adjustment request, the pricing data from the data store for a selected drug benefit claim, wherein the adjustment request is associated with a known error in pricing of the selected drug benefit claim;
pre-processing the retrieved pricing data for machine learning;
executing a machine learning classifier to create one or more candidate models;
training the candidate models based on a training set of the retrieved pricing data;
selecting one of the trained candidate models that meets a predetermined accuracy threshold as a predictive model;
executing the predictive model in response to the adjustment request based on a testing set of the retrieved pricing data, wherein the predictive model identifies a first predicted set of drug benefit claims impacted by the known error;
causing a sample of the first predicted set of drug benefit claims to be repriced in adjustment of the known error;
training the predictive model as a function of the repriced sample to predict a second predicted set of drug benefit claims impacted by the known error; and
storing the predictive model in the model repository of the database service.

2. The system of claim 1 , wherein the modeling processor is further configured for causing the second predicted set of drug benefit claims to be repriced in adjustment of the known error.

3. The system of claim 1 , wherein the modeling processor is further configured for executing one or more machine learning algorithms to generate and train the candidate models.

4. The system of claim 3 , wherein the one or more machine learning algorithms comprise at least one of a decision tree classifier and a probabilistic classifier.

5. The system of claim 1 , wherein pre-processing the retrieved pricing data includes converting non-numeric pricing data into categorical columns of numeric characterizations.

6. The system of claim 5 , wherein pre-processing the retrieved pricing data further includes executing one or more feature reduction operations to remove at least one of the columns containing a null value before generating the predictive model.

7. The system of claim 5 , wherein pre-processing the retrieved pricing data further includes executing one or more feature reduction operations to remove at least one of the columns having a variance lower than a threshold.

8. The system of claim 1 , wherein pre-processing the retrieved pricing data includes splitting the pricing data into the training set and the testing set.

9. The system of claim 1 , wherein the modeling processor is further configured to select the sample of the first predicted set of drug benefit claims within a date range as a function of a date of the known error.

10. A method comprising:
generating, by a front end, an adjustment request in response to user input;
retrieving, in response to the adjustment request, pricing data for a selected drug benefit claim from a data store, wherein the data store stores pricing data for a plurality of drug benefit claims and wherein the adjustment request is associated with a known error in pricing of the selected drug benefit claim;
pre-processing the retrieved pricing data for machine learning;
executing, by a modeling processor, a machine learning classifier to create one or more candidate models;
training the candidate models based on a training set of the retrieved pricing data;
selecting one of the trained candidate models that meets a predetermined accuracy threshold as a predictive model;
executing, by the modeling processor, the predictive model in response to the adjustment request based on a testing set of the retrieved pricing data, wherein the predictive model identifies a first predicted set of drug benefit claims impacted by the known error;
causing a sample of the first predicted set of drug benefit claims to be repriced in adjustment of the known error;
training the predictive model as a function of the repriced sample to predict a second predicted set of drug benefit claims impacted by the known error; and
storing the predictive model in a model repository.

11. The method of claim 10 , further comprising causing the second predicted set of drug benefit claims to be repriced in adjustment of the known error.

12. The method of claim 10 , further comprising executing one or more machine learning algorithms to generate and train the predictive model.

13. The method of claim 12 , wherein the one or more machine learning algorithms comprise at least one of a decision tree classifier and a probabilistic classifier.

14. The method of claim 10 , wherein pre-processing the retrieved pricing data includes converting non-numeric pricing data into categorical columns of numeric characterizations.

15. The method of claim 14 , wherein pre-processing the retrieved pricing data further includes executing one or more feature reduction operations to remove at least one of the columns containing a null value before generating the predictive model.

16. The method of claim 14 , wherein pre-processing the retrieved pricing data further includes executing one or more feature reduction operations to remove at least one of the columns having a variance lower than a threshold.

17. The method of claim 10 , wherein pre-processing the retrieved pricing data includes splitting the pricing data into a training set and a testing set.

18. The method of claim 10 , wherein the modeling processor is further configured to select the sample of the first predicted set of drug benefit claims within a date range as a function of a date of the known error.

19. A machine learning system comprising:
a modeling processor coupled to a data store and a database service of a pharmacy benefits management system, the data store storing pricing data for a plurality of drug benefit claims, the database service comprising a model repository; and
a memory storing computer-executable instructions that, when executed by the modeling processor, configure the modeling processor for:
retrieving, in response to an adjustment request, the pricing data from the data store for a selected drug benefit claim, wherein the adjustment request is generated by a front end of the pharmacy benefits management system and is associated with a known error in pricing of the selected drug benefit claim;
pre-processing the retrieved pricing data for machine learning;
executing a machine learning classifier to create one or more candidate models;
training the candidate models based on a training set of the retrieved pricing data;
selecting one of the trained candidate models that meets a predetermined accuracy threshold as a predictive model;
executing the predictive model in response to the adjustment request based on a testing set of the retrieved pricing data, wherein the predictive model

Assignees

• Express Scripts Strategic Dev Inc

Inventors

• Ciliberti John
• Gollapudi Amit
• Ball Richard

Classifications

• G06F18/24323

  Tree-organised classifiers · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06Q40/08 Primary

  Insurance · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F17/18

  for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)} · CPC title